Posts in 'blog' – Page 2

Testing Randomness in Python

In a side project I've been working on recently I've needed to write unit tests for some methods that return a non-deterministic random result.

In this case there was a method that would simulate an event happening based on some kind of probability factor. Eg if the factor was 0.2 then the event would happen on average every fifth time the method was called. It is pretty hard to unit test the outcome of that as the tests will fail 4/5 of the time. I needed to test that if I called that method 100 times with a factor of 0.2 that the event would happen roughly 20 times. Although of course sometimes it would be 19 times, sometimes 21. How do you test that?

After lots of messing about trying to Mock the random number generator and the random functions I discovered a much simpler approach. The random number generator on a computer is never actually truly random. It is a pseudo-random number generator (PRNG) which is in turn seeded with some 'randomness', eg the interrupts from the disk and network controller. But given the same seed the PNRG will always produce the same sequence of numbers when called.

So all I have to do is seed the PNRG with a known value before my test and the test outcome will always be deterministic.

An example below in Python, but most other languages have a similar seed() function.

import random

class MyClass:
  samples = ['a', 'b', 'c', 'd', 'e']

  def pickone():
    return random.choice(samples)

class MyTests():

  def testPickOne():
    picker = MyClass()
    choice = picker.pickone()
    self.assertEqual(choice, 'c')

if __class__ == '__main__':

You don't know what the outcome is going to be the first time you run the test. So you are likely to have to run it and then substitute in the returned value.

Actually in Python we can make this even nicer by using a decorator. In this way we can write a decorator that stores the random seed before we reset it and restores if afterward:

from decorator import decorator
import random

def fix_random(f, *args, **kw):
    state = random.getstate()
    res = f(*args, **kw)
    return res

or even as a context manager:

import random
class FixRandom():

    def __enter__(self):
        self.state = random.getstate()

    def __exit__(self, *args):

and can then be used:

print random.randint(0,100)

for i in range(10):
  with FixRandom():
    print random.randint(0,100)

print random.randint(0,100)

As you can see, produces the same ‘random’ number when called within the context, but an actual random one outside.


Finding Provisioning Profiles by Name Rather than UUID

Despite my previous thoughts, you can actually use Xcode 8’s find-by-name functionality for manual signing on a CI setup.

Duplicate Keychain Entries in macOS Sierra

I upgraded a build server to macOS Sierra and the keychain unlocking stopped working, here is how I fixed it.

Lockable Resources Jenkins 2.0 pipeline builds

If you want to increase concurrency in your Jenkins pipeline builds, but need to ensure that certain resources are not used concurrently, then you can use lockable resources

Changes to Keychains in macOS Sierra

I upgraded a build server to macOS Sierra and the keychain unlocking stopped working, here is how I fixed it.

Wiping the workspace in Jenkins 2.0 pipeline builds

Instead of the 'wipe workspace' checkbox that used to be in Jenkins, you can now use deleteDir() to clear the workspace before building.

WOTRA Trademark Scam letter

I filed for a trademark for my company name, shortly after it was accepted I received this scam letter in the post, quite a legitimate phishing attempt.

Using Xcode 8’s New Automatic Signing with Jenkins and Fastlane

Xcode 8 brings with it a new automatic code signing system. It is meant to make life a lot easier for developers, but needs a bit of work to get working with headless CI systems like Fastlane and Jenkins.

Uploading Git Changelog to Fabric Beta for Android Gradle Builds in Jenkins

The git changelog is not exposed as a variable in Jenkins for pipeline builds to use. This is how we got it and send it via Gradle to Fabric Beta when we distribute our automated builds

Global Build Numbers in Jenkins Multibranch Pipeline Builds

We wanted to have build numbers that were unique and incremental across all of our build jobs. Here is how I did it with a small python microservice.